mm/truncate.c

   1 /*
   2  * mm/truncate.c - code for taking down pages from address_spaces
   3  *
   4  * Copyright (C) 2002, Linus Torvalds
   5  *
   6  * 10Sep2002    akpm@zip.com.au
   7  *              Initial version.
   8  */
   9
  10 #include <linux/kernel.h>
  11 #include <linux/mm.h>
  12 #include <linux/swap.h>
  13 #include <linux/module.h>
  14 #include <linux/pagemap.h>
  15 #include <linux/pagevec.h>
  16 #include <linux/task_io_accounting_ops.h>
  17 #include <linux/buffer_head.h>  /* grr. try_to_release_page,
  18                                    do_invalidatepage */
  19
  20
  21 /**
  22  * do_invalidatepage - invalidate part of all of a page
  23  * @page: the page which is affected
  24  * @offset: the index of the truncation point
  25  *
  26  * do_invalidatepage() is called when all or part of the page has become
  27  * invalidated by a truncate operation.
  28  *
  29  * do_invalidatepage() does not have to release all buffers, but it must
  30  * ensure that no dirty buffer is left outside @offset and that no I/O
  31  * is underway against any of the blocks which are outside the truncation
  32  * point.  Because the caller is about to free (and possibly reuse) those
  33  * blocks on-disk.
  34  */
  35 void do_invalidatepage(struct page *page, unsigned long offset)
  36 {
  37         void (*invalidatepage)(struct page *, unsigned long);
  38         invalidatepage = page->mapping->a_ops->invalidatepage;
  39 #ifdef CONFIG_BLOCK
  40         if (!invalidatepage)
  41                 invalidatepage = block_invalidatepage;
  42 #endif
  43         if (invalidatepage)
  44                 (*invalidatepage)(page, offset);
  45 }
  46
  47 static inline void truncate_partial_page(struct page *page, unsigned partial)
  48 {
  49         memclear_highpage_flush(page, partial, PAGE_CACHE_SIZE-partial);
  50         if (PagePrivate(page))
  51                 do_invalidatepage(page, partial);
  52 }
  53
  54 void cancel_dirty_page(struct page *page, unsigned int account_size)
  55 {
  56         /* If we're cancelling the page, it had better not be mapped any more */
  57         if (page_mapped(page)) {
  58                 static unsigned int warncount;
  59
  60                 WARN_ON(++warncount < 5);
  61         }
  62
  63         if (TestClearPageDirty(page) && account_size &&
  64                         mapping_cap_account_dirty(page->mapping)) {
  65                 dec_zone_page_state(page, NR_FILE_DIRTY);
  66                 task_io_account_cancelled_write(account_size);
  67         }
  68 }
  69
  70 /*
  71  * If truncate cannot remove the fs-private metadata from the page, the page
  72  * becomes anonymous.  It will be left on the LRU and may even be mapped into
  73  * user pagetables if we're racing with filemap_nopage().
  74  *
  75  * We need to bale out if page->mapping is no longer equal to the original
  76  * mapping.  This happens a) when the VM reclaimed the page while we waited on
  77  * its lock, b) when a concurrent invalidate_inode_pages got there first and
  78  * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
  79  */
  80 static void
  81 truncate_complete_page(struct address_space *mapping, struct page *page)
  82 {
  83         if (page->mapping != mapping)
  84                 return;
  85
  86         cancel_dirty_page(page, PAGE_CACHE_SIZE);
  87
  88         if (PagePrivate(page))
  89                 do_invalidatepage(page, 0);
  90
  91         ClearPageUptodate(page);
  92         ClearPageMappedToDisk(page);
  93         remove_from_page_cache(page);
  94         page_cache_release(page);       /* pagecache ref */
  95 }
  96
  97 /*
  98  * This is for invalidate_inode_pages().  That function can be called at
  99  * any time, and is not supposed to throw away dirty pages.  But pages can
 100  * be marked dirty at any time too, so use remove_mapping which safely
 101  * discards clean, unused pages.
 102  *
 103  * Returns non-zero if the page was successfully invalidated.
 104  */
 105 static int
 106 invalidate_complete_page(struct address_space *mapping, struct page *page)
 107 {
 108         int ret;
 109
 110         if (page->mapping != mapping)
 111                 return 0;
 112
 113         if (PagePrivate(page) && !try_to_release_page(page, 0))
 114                 return 0;
 115
 116         ret = remove_mapping(mapping, page);
 117
 118         return ret;
 119 }
 120
 121 /**
 122  * truncate_inode_pages - truncate range of pages specified by start and
 123  * end byte offsets
 124  * @mapping: mapping to truncate
 125  * @lstart: offset from which to truncate
 126  * @lend: offset to which to truncate
 127  *
 128  * Truncate the page cache, removing the pages that are between
 129  * specified offsets (and zeroing out partial page
 130  * (if lstart is not page aligned)).
 131  *
 132  * Truncate takes two passes - the first pass is nonblocking.  It will not
 133  * block on page locks and it will not block on writeback.  The second pass
 134  * will wait.  This is to prevent as much IO as possible in the affected region.
 135  * The first pass will remove most pages, so the search cost of the second pass
 136  * is low.
 137  *
 138  * When looking at page->index outside the page lock we need to be careful to
 139  * copy it into a local to avoid races (it could change at any time).
 140  *
 141  * We pass down the cache-hot hint to the page freeing code.  Even if the
 142  * mapping is large, it is probably the case that the final pages are the most
 143  * recently touched, and freeing happens in ascending file offset order.
 144  */
 145 void truncate_inode_pages_range(struct address_space *mapping,
 146                                 loff_t lstart, loff_t lend)
 147 {
 148         const pgoff_t start = (lstart + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
 149         pgoff_t end;
 150         const unsigned partial = lstart & (PAGE_CACHE_SIZE - 1);
 151         struct pagevec pvec;
 152         pgoff_t next;
 153         int i;
 154
 155         if (mapping->nrpages == 0)
 156                 return;
 157
 158         BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1));
 159         end = (lend >> PAGE_CACHE_SHIFT);
 160
 161         pagevec_init(&pvec, 0);
 162         next = start;
 163         while (next <= end &&
 164                pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
 165                 for (i = 0; i < pagevec_count(&pvec); i++) {
 166                         struct page *page = pvec.pages[i];
 167                         pgoff_t page_index = page->index;
 168
 169                         if (page_index > end) {
 170                                 next = page_index;
 171                                 break;
 172                         }
 173
 174                         if (page_index > next)
 175                                 next = page_index;
 176                         next++;
 177                         if (TestSetPageLocked(page))
 178                                 continue;
 179                         if (PageWriteback(page)) {
 180                                 unlock_page(page);
 181                                 continue;
 182                         }
 183                         truncate_complete_page(mapping, page);
 184                         unlock_page(page);
 185                 }
 186                 pagevec_release(&pvec);
 187                 cond_resched();
 188         }
 189
 190         if (partial) {
 191                 struct page *page = find_lock_page(mapping, start - 1);
 192                 if (page) {
 193                         wait_on_page_writeback(page);
 194                         truncate_partial_page(page, partial);
 195                         unlock_page(page);
 196                         page_cache_release(page);
 197                 }
 198         }
 199
 200         next = start;
 201         for ( ; ; ) {
 202                 cond_resched();
 203                 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
 204                         if (next == start)
 205                                 break;
 206                         next = start;
 207                         continue;
 208                 }
 209                 if (pvec.pages[0]->index > end) {
 210                         pagevec_release(&pvec);
 211                         break;
 212                 }
 213                 for (i = 0; i < pagevec_count(&pvec); i++) {
 214                         struct page *page = pvec.pages[i];
 215
 216                         if (page->index > end)
 217                                 break;
 218                         lock_page(page);
 219                         wait_on_page_writeback(page);
 220                         if (page->index > next)
 221                                 next = page->index;
 222                         next++;
 223                         truncate_complete_page(mapping, page);
 224                         unlock_page(page);
 225                 }
 226                 pagevec_release(&pvec);
 227         }
 228 }
 229 EXPORT_SYMBOL(truncate_inode_pages_range);
 230
 231 /**
 232  * truncate_inode_pages - truncate *all* the pages from an offset
 233  * @mapping: mapping to truncate
 234  * @lstart: offset from which to truncate
 235  *
 236  * Called under (and serialised by) inode->i_mutex.
 237  */
 238 void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
 239 {
 240         truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
 241 }
 242 EXPORT_SYMBOL(truncate_inode_pages);
 243
 244 /**
 245  * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
 246  * @mapping: the address_space which holds the pages to invalidate
 247  * @start: the offset 'from' which to invalidate
 248  * @end: the offset 'to' which to invalidate (inclusive)
 249  *
 250  * This function only removes the unlocked pages, if you want to
 251  * remove all the pages of one inode, you must call truncate_inode_pages.
 252  *
 253  * invalidate_mapping_pages() will not block on IO activity. It will not
 254  * invalidate pages which are dirty, locked, under writeback or mapped into
 255  * pagetables.
 256  */
 257 unsigned long invalidate_mapping_pages(struct address_space *mapping,
 258                                 pgoff_t start, pgoff_t end)
 259 {
 260         struct pagevec pvec;
 261         pgoff_t next = start;
 262         unsigned long ret = 0;
 263         int i;
 264
 265         pagevec_init(&pvec, 0);
 266         while (next <= end &&
 267                         pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
 268                 for (i = 0; i < pagevec_count(&pvec); i++) {
 269                         struct page *page = pvec.pages[i];
 270                         pgoff_t index;
 271                         int lock_failed;
 272
 273                         lock_failed = TestSetPageLocked(page);
 274
 275                         /*
 276                          * We really shouldn't be looking at the ->index of an
 277                          * unlocked page.  But we're not allowed to lock these
 278                          * pages.  So we rely upon nobody altering the ->index
 279                          * of this (pinned-by-us) page.
 280                          */
 281                         index = page->index;
 282                         if (index > next)
 283                                 next = index;
 284                         next++;
 285                         if (lock_failed)
 286                                 continue;
 287
 288                         if (PageDirty(page) || PageWriteback(page))
 289                                 goto unlock;
 290                         if (page_mapped(page))
 291                                 goto unlock;
 292                         ret += invalidate_complete_page(mapping, page);
 293 unlock:
 294                         unlock_page(page);
 295                         if (next > end)
 296                                 break;
 297                 }
 298                 pagevec_release(&pvec);
 299         }
 300         return ret;
 301 }
 302
 303 unsigned long invalidate_inode_pages(struct address_space *mapping)
 304 {
 305         return invalidate_mapping_pages(mapping, 0, ~0UL);
 306 }
 307 EXPORT_SYMBOL(invalidate_inode_pages);
 308
 309 /*
 310  * This is like invalidate_complete_page(), except it ignores the page's
 311  * refcount.  We do this because invalidate_inode_pages2() needs stronger
 312  * invalidation guarantees, and cannot afford to leave pages behind because
 313  * shrink_list() has a temp ref on them, or because they're transiently sitting
 314  * in the lru_cache_add() pagevecs.
 315  */
 316 static int
 317 invalidate_complete_page2(struct address_space *mapping, struct page *page)
 318 {
 319         if (page->mapping != mapping)
 320                 return 0;
 321
 322         if (PagePrivate(page) && !try_to_release_page(page, GFP_KERNEL))
 323                 return 0;
 324
 325         write_lock_irq(&mapping->tree_lock);
 326         if (PageDirty(page))
 327                 goto failed;
 328
 329         BUG_ON(PagePrivate(page));
 330         __remove_from_page_cache(page);
 331         write_unlock_irq(&mapping->tree_lock);
 332         ClearPageUptodate(page);
 333         page_cache_release(page);       /* pagecache ref */
 334         return 1;
 335 failed:
 336         write_unlock_irq(&mapping->tree_lock);
 337         return 0;
 338 }
 339
 340 /**
 341  * invalidate_inode_pages2_range - remove range of pages from an address_space
 342  * @mapping: the address_space
 343  * @start: the page offset 'from' which to invalidate
 344  * @end: the page offset 'to' which to invalidate (inclusive)
 345  *
 346  * Any pages which are found to be mapped into pagetables are unmapped prior to
 347  * invalidation.
 348  *
 349  * Returns -EIO if any pages could not be invalidated.
 350  */
 351 int invalidate_inode_pages2_range(struct address_space *mapping,
 352                                   pgoff_t start, pgoff_t end)
 353 {
 354         struct pagevec pvec;
 355         pgoff_t next;
 356         int i;
 357         int ret = 0;
 358         int did_range_unmap = 0;
 359         int wrapped = 0;
 360
 361         pagevec_init(&pvec, 0);
 362         next = start;
 363         while (next <= end && !ret && !wrapped &&
 364                 pagevec_lookup(&pvec, mapping, next,
 365                         min(end - next, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
 366                 for (i = 0; !ret && i < pagevec_count(&pvec); i++) {
 367                         struct page *page = pvec.pages[i];
 368                         pgoff_t page_index;
 369
 370                         lock_page(page);
 371                         if (page->mapping != mapping) {
 372                                 unlock_page(page);
 373                                 continue;
 374                         }
 375                         page_index = page->index;
 376                         next = page_index + 1;
 377                         if (next == 0)
 378                                 wrapped = 1;
 379                         if (page_index > end) {
 380                                 unlock_page(page);
 381                                 break;
 382                         }
 383                         wait_on_page_writeback(page);
 384                         while (page_mapped(page)) {
 385                                 if (!did_range_unmap) {
 386                                         /*
 387                                          * Zap the rest of the file in one hit.
 388                                          */
 389                                         unmap_mapping_range(mapping,
 390                                            (loff_t)page_index<<PAGE_CACHE_SHIFT,
 391                                            (loff_t)(end - page_index + 1)
 392                                                         << PAGE_CACHE_SHIFT,
 393                                             0);
 394                                         did_range_unmap = 1;
 395                                 } else {
 396                                         /*
 397                                          * Just zap this page
 398                                          */
 399                                         unmap_mapping_range(mapping,
 400                                           (loff_t)page_index<<PAGE_CACHE_SHIFT,
 401                                           PAGE_CACHE_SIZE, 0);
 402                                 }
 403                         }
 404                         if (!invalidate_complete_page2(mapping, page))
 405                                 ret = -EIO;
 406                         unlock_page(page);
 407                 }
 408                 pagevec_release(&pvec);
 409                 cond_resched();
 410         }
 411         WARN_ON_ONCE(ret);
 412         return ret;
 413 }
 414 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
 415
 416 /**
 417  * invalidate_inode_pages2 - remove all pages from an address_space
 418  * @mapping: the address_space
 419  *
 420  * Any pages which are found to be mapped into pagetables are unmapped prior to
 421  * invalidation.
 422  *
 423  * Returns -EIO if any pages could not be invalidated.
 424  */
 425 int invalidate_inode_pages2(struct address_space *mapping)
 426 {
 427         return invalidate_inode_pages2_range(mapping, 0, -1);
 428 }
 429 EXPORT_SYMBOL_GPL(invalidate_inode_pages2);